Statistical methods Topic on Variables .pdf

Lecture on statistical methods, with current topic on variables

1. Variables
27 October 2018
The PICO(TS) Model forResearch QuestionsPPatient, Population, or ProblemHospital-acquired infection
IIntervention, Prognostic factor, or ExposureHandwashing
CComparison, or Intervention (if appropriate)No handwashing; other solution; maskOOutcome you would like to measure or achieveReduced infection
TType of question you’re asking(Diagnosis, Etiology/Harm, Therapy, Prognosis, Prevention)SStudy type(Study design)
Variable
•Its value variesfrom one individual to another or within the same individual at different periods of time.•Types:–Qualitative–Quantitative•Discrete•Continuous
•Levels–Nominal–Ordinal–Interval–Ratio(chat noir)
Variables according to how they are expressed/measured•Qualitative variables–Variables whose categories are simply used as labelsto distinguish one group from another–Numerical representation of the categories are for labeling/coding and not for comparison (greater or less)–E.g., sex, religion, place of residence, disease status

2. Variables according to how they are expressed/measured•Quantitative variables–Values indicate a quantity or amount and can be expressed numerically–Values can be arranged according to magnitude–E.g., age, height, weight, blood pressure–May be either discreteor continuous
Types of quantitative variables
•Discrete–Can assume only integral values or whole numbers (finite number of values)–E.g., number of children in the family, number of beds in the hospital–Counts, ratios•Continuous–Can attain any value including fractions or decimals (possible values fall on a continuum)–E.g., height, weight–Includes proportions, rates
Variables according to levels of measurement•Nominal–A classificatory scale where the categories are used as labels only (does not represent quantity)–Number or names which represent a set of mutually exclusive and exhaustivecategoriesto which individuals or objects (attributes) may be assigned–Most discrete variables are nominal–E.g., sex (M, F), race, blood groups, seatbelts in car, psychiatric diagnosis, patient ID no.
Variables according to levels of measurement•Ordinal–Same characteristics as the nominalscale–Additional feature: categoriescan be orderedor ranked; however, the distance between the two categories cannot be clearly quantified–E.g., Likertscales, age groups (infant, child, teenager, adult), psychosocial scales (strongly disagree, disagree, agree, strongly agree)

3. Variables according to levels of measurement•Interval–Same characteristics as the ordinalscales–Additional feature: distancesbetween all adjacent classes are equal–Conceptually, these scales are infinite, in that they have neither beginning nor ending–Zero point is arbitraryand does not mean absence of the characteristic–E.g., temperature, IQ–Can be quantitative continuous
Variables according to levels of measurement•Ratio–Same characteristics as for ordinalscales–A meaningfulzero point exists–Ratio of two numbers can be meaningfully computed and interpreted–E.g., weight, blood pressure, height, doctor visits, number of DMF teeth–Can be quantitative continuous
Levels of Measurement
Information increases; progressively more precise mathematically
Variables according to causal relationships•Independent variable–Variable under investigation, hypothesized to have an effect on the outcome–Presumed cause (exposure)in an experimental study; under the experimenter’s control•Dependent variable–Outcome of interest in the investigation–Presumed effectin an experimental study; value dependent upon the independent variable

4. Variables according to causal relationships
Independent variableDependent variable
Variables according to its relationship with the E and O•Confounding variable–Extraneousvariable that distortsthe relationship between exposure and disease àunder/overestimation of effect–Associated with the exposure but is not a consequence of the exposure–A risk factor for the study disease
A scenario:Another scenario:

5. Yet another scenario:
Amount of salaryWork productivity
What if they are already hard workers to begin with? They really like what they do? Don’t want to leave the office first? Passionate? Motivated? Competitive? High pay is incidental. How do we know what the cause of their increased productivity is?
Yet another scenario:
Amount of salaryWork productivity
PersonalityMotivationCompetition
Variables according to its relationship with the E and O
•Intermediate variable–Variable along the causal pathway between exposure and outcome–Causes variation in the dependent variable–Is itself caused to vary by the independent variable
Confounding vsEffect modification•Confounding–distortion of the association between an exposure and an outcome that occurs when the study groups differ with respect to other factors that influence the outcome•Effect measure modification –occurs when the magnitude of the effect of the primary exposure on an outcome (i.e., the association) differs depending on the level of a third variable

6. Confounding
DiedSurvivedTOTALHospital A7320372110Hospital B16784800TOTAL8928212910RR = 1.73 (95% CI:1.01, 2.95)
DiedSurvivedTOTALHospital A6514431508Hospital B8192800TOTAL7316351708RR = 1.08 (95% CI:0.52, 2.21)
DiedSurvivedTOTALHospital A8594602Hospital B8592600TOTAL1611861202RR = 1.00 (95% CI:0.38, 2.64)
Patients severely illPatients notseverely ill
Accounting for the confounding variable:“severity of illness”
Confounding
Crude ORAdjustedORComments2.5(95% CI:1.5, 4.2)1.0Confounders distort the E–D relationship to the extent that they completelyexplain the crude OR2.5(95% CI:1.5, 4.2)2.0(95% CI: 1.1, 3.4)Confoundersmodestly distort E–D relationship; adjusted OR still statistically significant2.5(95% CI:1.5, 4.2)3.5(95% CI: 2.2, 5.4)Confounder distorts the relationshipbutin the opposite direction (could be a protective factor rather than a risk factor*)0.40(95% CI:0.15, 0.75)0.65(95%CI: 0.40, 1.2)Confoundersmodestly distort E–D relationship; adjusted OR still not statistically significant
Effect measure modification(New drug and HDL increase)
Source: Boston University School of Public Health
1
A statistically significant HDL-C increase with the new drug was expected.Is there another variable masking the effect of the treatment?
Effect measure modification(New drug and HDL increase)
Source: Boston University School of Public Health

7. Effect measure modification(Hospitalization for MVA)
(95% CI: 1.32, 1.56)
(95% CI: 1.62, 2.00)
(95% CI: 0.80, 1.06)Source: Boston University School of Public Health
Confounding vs Effect modification•Confounding is a distortionof the true association caused by an imbalanceof some other risk factor.•Effect modification is a biological phenomenonthat should be described.–Pooled data can be misleading.–Stratum-specificmeasures of association estimates should be reported separately.–Allows for interaction.
Dependent variableIndependent variable
?Confounding variable/s
?Intermediate variable/s
Effect modifiers?
Conceptual vsOperational Definitions•Conceptual definition of a variable–Nominal definition; dictionary definition–Uses literal terms to specify the qualities of a variable–Tells you what the concept of that variable means•Operational definition of a variable–Specifies the procedures and criteria for taking a measurement of that variable–Tells you how to measure that variable

8. Conceptual vsOperational DefinitionsConceptual vsOperational Definitions
VariableConceptual DefinitionOperationalDefinitionWeightThe heaviness of an objectHeaviness measured in kilogramsAgeAmountof time during which a person has livedLength of time from birth measured in months or age in yearsat last birthdayHemoglobin levelLevelof thered proteinresponsible for transporting oxygen in the blood of vertebrates
Hemoglobin concentration in capillaryblood, measure in mg/dl using a hemoglobinometerBMIA measurefor human body shape based on an individual’s mass and height
Theweight in kilograms divided by the square of the height in meters (kg/m2)
References
•Biostatistics 201 Lecture Notes (2013), Department of Epidemiology and Biostatistics, College of Public Health, University of the Philippines Manila.•Boston University School of Public Health. Confounding and Effect Measure Modification. Accessed at http://sphweb.bumc.bu.edu/otlt/MPH-Modules/BS/BS704-EP713_Confounding-EM/BS704-EP713_Confounding-EM8.html(21 Oct 2018)•Daniel WW (2009). Biostatistics: A Foundation for Analysis in the Health Sciences, 9thedition. Wiley & Sons, Inc., USA.•Evans JD (1996). Straightforward Statistics for the Behavioral Sciences. Brooks/Cole Publishing, Pacific Grove, CA, USA.•Katz DL, Elmore JG, Wild DMG (2014). Jekel’sEpidemiology, Biostatistics, Preventive Medicine, and Public Health, 4thedition. Elsevier Saunders, Philadelphia, PA, USA.•Mendoza OM, Borja MP (2010). Foundations of Statistical Analysis for the Health Sciences, Department of Epidemiology and Biostatistics, College of Public Health, University of the Philippines Manila.•MomeniA, Pincus M, LibienJ (2018). Introduction to Statistical Methods in Pathology. Springer International Publishing.•Riffenburgh RH (2012). Statistics in Medicine, 3rdedition. Elsevier, Inc., USA.